The US government has forced Anthropic to block public access to its latest AI model just days after release, citing a potential security vulnerability – and the company’s chief executive is now heading to Washington for a high-stakes showdown.
On Friday, Anthropic abruptly shut down access to Fable 5, a version of its Claude Mythos AI that the company itself had warned was “too powerful” for the public. The move came after the US government prohibited the firm from allowing any foreign national to use the technology, according to two people familiar with the matter. Anthropic said it had only received “verbal evidence” of the so-called jailbreak vulnerability.
“US government blocks Anthropic's AI tool Fable 5 over security fears; CEO to meet Commerce Secretary Monday.”
The sudden intervention sets the stage for a meeting on Monday in Washington DC between Anthropic CEO Dario Amodei and Secretary of Commerce Howard Lutnick. The meeting is expected to include more documentation of the alleged issue, one source said.
Anthropic had released Fable 5 on Tuesday, describing it as a version with extra safeguards, while a more powerful variant, Mythos 5, remained limited to a select group of organizations. Both represent a new iteration of Claude Mythos, which caused a stir when it was first previewed privately in April to a small number of groups, including departments inside the US government itself.
“Fable’s capabilities exceed those of any model we’ve ever made generally available,” Anthropic said at launch, adding that “releasing a model this capable comes with risks.” The company said the tool could work “unattended” on human commands for longer periods than any previous Claude model.
But within days, the US government said it had “become aware” of a potential jailbreak – an opening that could allow the AI to be used beyond its intended purpose. Anthropic pushed back, insisting it had only verbal evidence.
The standoff is the latest in a strained relationship between Anthropic and the US government. The company is already suing the Department of Defense over restrictions on how its models can be used. However, tensions had appeared to ease after a “productive” meeting with White House officials a few weeks ago.
Despite the block, dozens of tech leaders – including security staff from Nvidia, Zoom, and Mercedes-Benz, as well as former US government and Google security officials – have signed an open letter urging the government to allow Anthropic to release the models publicly.
The controversy has drawn attention from beyond the tech world. Canadian Finance Minister François-Philippe Champagne told the BBC in April that the attention on Mythos was warranted because “it’s the unknown, unknown.” Meanwhile, the roughly 150 groups that had previewed Mythos have now been given access to Mythos 5 without limitations on cybersecurity or biology, depending on their use. Those groups have already reported finding more than 10,000 critical security flaws in their systems using the tool.
Anthropic’s valuation has neared $1tn, and the company is expected to go public soon. Co-founder Jack Clark told BBC Newsnight last week that the rapid expansion of AI capabilities meant “you want the option to be able to take your foot off the gas and put your foot on the brake.”
It remains unclear whether, after Monday’s meeting, Fable 5 and Mythos 5 will ever be accessible to the public again. A White House spokesperson declined to comment.